PulseAugur
EN
LIVE 07:00:25
中文(ZH) 全球首个人形机器人通用小脑来了!全球最大规模2万小时人类动作数据,实现零样本泛化

Humanoid robot 'cerebellum' model trained on 20B frames, enters GPT era · 1 source tracked

Galaxy General Robotics has released AstraBrain-WBC 0.5, a foundational model for the real-time whole-body control of humanoid robots. This model utilizes a Transformer architecture, similar to GPT, and is trained on approximately 20 billion frames of human motion data, making it the largest dataset of its kind for humanoid robot control. The development signifies a shift towards a "foundation model" era for robot motion, enabling enhanced generalization, stability, and adaptability in complex tasks. AI

IMPACT This model's GPT-like architecture and massive training data could accelerate the development of more capable and adaptable humanoid robots.

RANK_REASON New foundational model release for humanoid robot control from a robotics lab. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on 量子位 (QbitAI) →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Humanoid robot 'cerebellum' model trained on 20B frames, enters GPT era · 1 source tracked

COVERAGE [1]

  1. 量子位 (QbitAI) TIER_1 中文(ZH) · 衡宇 ·

    The world's first general-purpose cerebellum for humanoid robots is here! The world's largest-scale 20,000-hour human motion dataset achieves zero-shot generalization.

    人形机器人正式迈入“GPT时代”